Is this bug or not?

[Steven Schveighoffer]

On Fri, 06 May 2011 08:49:44 -0400, Jason House  
<jason.james.house at gmail.com> wrote:

> Steven Schveighoffer Wrote:
>
>> On Fri, 06 May 2011 03:39:31 -0400, Mariusz Gliwiński
>> <alienballance at gmail.com> wrote:
>>
>> > Is this bug, or is it as supposed to be? I'm not sure...
>> > <code>
>> > interface TmpI(T) {
>> >      void func(T);
>> > }
>> >
>> > class TmpC {}
>> > class TmpC2 : TmpC {}
>> >
>> > class Tmp : TmpI!TmpC2 {
>> >      void func(TmpI!TmpC) {};
>>
>> Here is one problem, TmpI is instantiated with TmpC2 as the func
>> parameter, yet you use TmpI!TmpC as your func parameter.  The func
>> parameter should be TmpC2.  I'm surprised this part would compile  
>> without
>> complaining that you didn't implement func.  But it probably would fail  
>> if
>> it got to that check.
>
>
> This part is fine. When extending a class or interface, you are free to  
> widen the input and narrow the output. Users of the interface can pass  
> in TmpC2 (it's a valid TmpC)

But look at TmpI, if T is TmpC2, then func's param is TmpC2.  This can't  
be implemented with TmpI!TmpC.

Let's take templates out of it, and look at what is generated:

interface TmpI1 // in original exmample, this is TmpI!TmpC2
{
    void func(TmpC2);
}

interface TmpI2 // in original example, this is TmpI!TmpC
{
    void func(TmpC);
}

class Tmp : TmpI1
{
    void func(TmpI2) {}
}

There is no way TmpI2 casts implicitly to or from TmpC2, they are  
unrelated types.

-Steve